From the Classroom Floor: Code AI and the Case for Teaching Coding WITH AI

I spent forty-five minutes actively rethinking the most common criticism of vibe coding.

The criticism goes something like this: if students are using AI to generate their code, they are not actually learning to code. They are learning to prompt. They are outsourcing the thinking to a machine and calling it programming education, and the result is a generation of students who can describe what they want a computer to do but have no idea how computers actually do it.

It is a reasonable criticism. I have heard it from colleagues, from administrators, from computer science educators who have spent careers teaching syntax and logic and the specific discipline of making a machine do exactly what you tell it to do. I do not dismiss it. I think it deserves a genuine answer.

Code AI is that answer.

What Used to Be Code.org

If you have been in education for any length of time, you know Code.org. The nonprofit that brought the Hour of Code to schools worldwide. The organization that made the argument, convincingly and at scale, that every student deserves access to computer science education regardless of background, zip code, or prior experience.

They recently rebranded to Code AI. That is not a cosmetic change. It is a deliberate signal about where they believe computing education needs to go: not away from coding, but toward coding in the age of AI, which is a fundamentally different skill set than coding in the age of typing syntax into a terminal and hoping for the best.

At ISTE 2026, they had a classroom. A real one, set up near the expo, running demonstrations of their curriculum with interested educators. I sat down as a student for forty-five minutes and built a website.

Not with a template. Not by dragging and dropping pre-built components. I built a website with HTML, with structure, with real decisions about what the code was doing and why, while collaborating with an AI partner and a curricular package that explained every step to me along the way!

The Socratic Bot

Here is what I expected the AI collaboration to look like: I describe what I want, the AI generates the code, I paste it in, something appears on the screen, I feel vaguely accomplished without understanding what just happened.

That is not what happened.

The AI in Code AI's lesson functions more like a Socratic tutor than a code generator. (In the earlier lessons, we did not experience; students worked to develop HTML shells alongside the assistance of their AI thought partner) When I needed to do anything, it would make sure I was adding the correct files. That the skeletons I had selected matched up with what I wanted my end product to do. It was working truly as a partner, instead of a program carrying out a task.

By the end of forty-five minutes, I had a website. But more importantly, I understood — in a way I did not fully before — why the website worked. Not because I had memorized syntax. Because the AI had refused to let me skip the thinking.

The Vibe Coding Response

I teach a series called Vibe Coding for Educators. I presented sessions at TCEA and ISTE, arguing that AI has fundamentally lowered the barrier to building classroom tools and that teachers do not need to know how to code to create something functional and useful. I believe all of that, and I am not walking any of it back.

But here is what Code AI showed me: the criticism of vibe coding is not wrong about the risk. It is wrong about the solution.

The risk is real, if students use AI to generate code without understanding what the code is doing, they are not developing computational thinking. They are developing prompting skills, which are valuable, but they are not the same thing. The question is whether the answer to that risk is removing AI from the coding process or designing the AI's role in the coding process so that understanding is built in rather than bypassed.

Code AI's answer is the second one. The AI is not a shortcut around the learning. It is a Socratic partner that keeps redirecting you back to the learning every time you try to skip it.

That is not vibe coding. That is something more interesting, coding education that uses AI the same way DeskPad uses AI. Not to do the work. To make you do it better.

Why This Matters for Every Teacher, Not Just CS Teachers

I am a Social Sciences teacher. I am not teaching an HTML unit in my US History class (though this session convinced me that I could). But I walked out of that forty-five-minute session thinking about something bigger than coding.

The pedagogical model Code AI demonstrated, a Socratic AI partner that builds understanding by refusing to skip steps, is transferable to any subject where students are tempted to use AI to get to the answer without going through the thinking. Which is every subject. Which is every classroom.

The solution Code AI is exploring at the intersection of programming education and AI is the same question DeskPad is exploring about writing and the same issue every thoughtful educator is trying to answer right now: how do we design the AI's role in learning so that it builds capacity rather than replacing it?

A forty-five-minute session on the ISTE expo floor is not a complete answer to that. But it is one of the clearest demonstrations I saw all week of what the answer might look like in practice.

🔗 https://code.org